A few months ago, Western Digital offices were living through a real tumult. The reasons for this worry were quite serious: the memory chips suppliers seem to have connived with each other to stop producing 16Mbit SDRAM chips and shifted to the production of more profitable and up-to-date 64-256Mbit chips. The 16Mbit chips Western Digital had in stock kept dwindling before their very eyes, since WD hard disk drives enjoyed pretty high popularity. Besides, WD had to avoid letting down one of its major OEM-partners, Microsoft, which was filling its stocks with X-Box consoles before the retail sales started.

In despair, Western Digital purchasing office signed fettering agreements about the shipments of the memory chips available in the market then. However, in order to make up for the price difference between the new chips and the formerly used 16Mbit ones, the faster and more profitable HDDs than the Low-End solutions acquired memory modules of greater capacity.

Then an even more incredible thing happened. By an oversight of one of WD's software developers, the hard drive DSP-processor could now access not only 2MB, but the entire 8MB memory. And which is also unbelievable, the HDD with this "crazy" DSP-processor turned out evidently faster than its counterparts featuring memory modules of lower capacity.

Luckily, WD decided not to conceal this fact, but to make some money on it. The hard disk drive model with larger cache-buffer was called "Special Edition" and started selling at a "special" price. The purchasing office and the software guy got a significant wage raise, and the competitors started whipping the cat and keep doing it still.

Anyway, enough joking :) Let's get down to something more serious.

For the last couple of years, almost all the parts of the hard disk drives underwent some changes. However, strange as it might seem, but the cache-buffer size, didn't increase as much as the data density, for instance. Even though the memory microchips capacity has been growing quite a lot lately.

We have already had a lot of opportunities to make sure that the cache-buffer size has direct influence on the HDD performance. For example, in the beginning of last year, WD Protege 200EB HDD with 2MB cache buffer defeated completely IBM DTLA 3050x0 with 512KB cache-buffer. All of you should also remember a not very successful experiment carried out by Quantum with their Lct20 model, which was equipped with only 128KB cache memory.

And now, finally, Western Digital made a significant move forward: they made the cache-buffer memory of their hard drives four times higher (so far only "special" models have it). Is it a marketing trick or a justified necessity?

Well, we are about to answer this question right away.

Closer Look

First of all, we would like to sincerely thank Western Digital for the WD 1200JB HDD they sent us. The beautiful box says in large letters: "120GB Special Edition":

And inside the box, there is … a hard disk drive!

To protect the drive against transportation damages it was packed into a special plastic bag.

Besides the hard drive, we also found a user's manual, a floppy disk with utilities, EDMA66/100 cable and a set of fastening screws. In other words, everything you may need to connect your new HDD to a PC.

And this is the drive itself:

   

Besides the "1200" there is hardly anything to catch your eye. Since the hard disk drives do not use any multicolour PCBs yet, the WD1200JB doesn't look any different from the WD1000BB HDD, for instance.

However, we couldn't help having a look at the "other side to the picture":

This is the reverse side of the hard drive PCB. Please have a look at the memory chip, which is located on the left of the bigger chip.

If you take a close look at it you will see the marking of a 64Mbit Nanya chip: NT56V6620C0T, which means that the cache buffer of the HDD makes 8MB (the same thing was proven by the IBM Feature Tool utility, actually).

The funny thing about this Nanya chip is that it is listed among the discontinued (EOL) products. We prefer not to imagine what it may mean :)

This time we will compare the performance of our today's hero, Western Digital WD1200JB with that of WD800BB, Maxtor D740X-6L (6L080J4), Seagate Barracuda ATA IV (ST3008021A), WD1000BB, IBM Deskstar 120GXP (IC35L080AVVA07-0).

Testbed and Methods

We tested the newcomer in the following test system:

• Intel Pentium III (Coppermine) 600MHz CPU;
• ASUS CUBX-E mainboard, bios 1007A;
• 2 x 128MB PC133 SDRAM by Hyundai;
• Matrox Millennium 4MB graphics card;
• Windows 98/Windows 2000 Pro.

In order to check the HDDs performance in different UDMA modes, we had to use the following controllers:

• UDMA33: the controller integrated into i440BX chipset;
• UDMA66: Promise Ultra66 controller;
• UDMA100: integrated Promise Ultra100 controller on ASUS CUBX-E mainboard.

For Promise controllers we used the drivers ver. 1.60 (build 33).

The disk drives were connected as Master-units to a separate IDE-channel. DMA support in Windows was enabled. We used FAT32 and NTFS file systems to format each of them as one logical drive of the maximum size with the default cluster. All the tests were run 4 times and then the average results were taken for the diagrams. The HDDs didn't rest for cooling down between the tests.

Here are the benchmarks used:

• Windows 98 WinBench 99 1.2
• Adaptec Threadmark 2.0
• Windows 2000 WinBench 99 1.2
• HDTach 2.61
• Intel IOMeter 1999.10.20

Performance

Average Access Time

Since both: HDTach and WinBench99 measure Average Access Time for reads, larger cache buffer of WD1200JB shouldn't have any influence on its performance here.

As we have expected, the average read rate from a random sector by WD1200JB is just the same as that of any other latest WD hard drive with the cache buffer of the regular size.

Sustained Linear Read Speed

As for the linear read speed, larger cache buffer also can't increase it. However, it is the first time we test a HDD with a 40GB platter. We wonder how Western Digital built it. Usually WD achieved higher data density per platter by increasing the number of tracks and this way the linear read speed remained pretty low. However, lately we could see the increase in the linear read speed of WD's hard drives, which took place together with the growth of the data density per platter. You can see it clearly on the diagram below:

As you see, WD's 33GB platter is faster than a 27GB platter, and a 40GB platter is faster than a 33GB one. If we compare the linear read speed of WD1200JB and IBM 120GXP, the IBM drive will be the winner here. A bit later we will have another look at the linear read graphs for WD1200JB solution and return to our discussion.

HDTach 2.61

What do we see? The results of all our testing participants are very close to one another, except those shown by Seagate Barracuda ATA IV. However, we didn't expect WD1200JB to show any superiority here over the rest of the racers, since the maximum read rate from the cache buffer doesn't depend on its size in any way.

However, the average read rate let WD1200JB dash forward leaving IBM Deskstar 120GXP 1MB/sec behind.

In the average write rate test, WD1200JB didn't prove too successful. It suddenly yielded even to WD1000BB, another WD's hard disk drive with lower per platter data density.

WinBench99 1.2 for Windows98

Now let's pass over to the most exciting part: WinBench tests. In fact, some of these tests have already become pretty outdates, so we are afraid that they may easily fit into the cache buffer of our today's hero without any problems.

Now have a look at the diagrams:

Well, WD1200JB won the round, though it didn't manage to get too far ahead of the competitors.

At the same time note that the results shown by the drive in UDMA33 mode have grown up significantly! We wonder what the High-End benchmarks will show:

Hm… Paradoxical as it might seem, but WD1200JB turned out the second yielding the first prize to IBM 120GXP. In UDMA33 mode however, the WD1200JB again leads the show with a great advantage over the rivals.

Adaptec Threadmark 2.0

If you remember, in our previous review we mentioned that IBM 120GXP stopped the chain of WD's victories in this tests. However, its success didn't last long: WD1200JB resumed it:

WD1200JB returned to the leading positions. Note that in UDMA33 mode it proved unbelievably fast.

WinBench99 1.2 for Win2000 (FAT32)

Now let's pass over to Win2000. maybe this operation system will help us discover the real fastness of WD1200JB?

DTR: Beginning

Graph

Graph

Graph

Well, the specialists have already estimated what it is worth, we suppose :) Now comes a very illustrative diagram:

Not bad, eh? WD1200JB HDD is 17% faster than IBM 120GXP, which used to be our leader and the advantage of WD1200JB over WD1000BB makes the good 24% already.

In High-End benchmarks WD1200JB performed pretty well, though it failed to get as far ahead of IBM 120GXP as in Business tests. Again, we would like to draw your attention to the fact that the drive showed very higher results in UDMA33 mode.

Returning to the discussion of WD1200JB linear read speed, please have a look at the graphs provided in the table above. The graphs do not look quite "linear". We tried really hard to linearize it, so the DTR: Beginning value obtained in this test seems to be not quite correct. However, we do have some idea of this number: 48-49MB/sec.

WinBench99 1.2 Win2000 (NTFS)

We have every right to call NTFS a favourite file system of WD hard disk drives. However, cut-throat competition between Wd1200JB and IBM 120GXP can continue here as well.

As we have expected, WD1200JB again won the first prize. This time it managed to outperform IBM 120GXP by 12.5% (which is quite a lot, we should say).

In High-End tests WD1200JB appeared 6% faster than IBM 120GXP, which is just the same difference as we saw in FAT32.

Well, judging by all the WinBench99 tests carried out in Win2000, we can call WD1200JB an absolute leader. Now it's high time for the hardest test: will our hero cope with it?

Intel IOMeter

We will begin with the three standard patterns:

For your convenience we collected all Total I/Os in a single table for all possible workloads:

In FileServer pattern in case of low workload (1-4 requests) the leadership belongs to Maxtor D740X-6L HDD and as the workload increases, IBM 120GXP takes the lead. WD1200JB won the third prize in this test, despite the large cache-buffer.

In WorkStation pattern, Maxtor was the leader only in case of the lowest workload, and in all other modes IBM solution proved the fastest. Strange as it might seem, but in this pattern Wd1200JB turned out slower than even WD1000BB HDD.

However, in DataBase pattern the new WD hard drive managed to regain its reputation: it headed the race in case on lower workloads, which are typical of desktop systems. As the workload grew up, IBM 120GXP dashed forward.

It is known that human brain remembers the images much better than numbers, therefore, we will repeat the already mentioned things, probably. But this time we suggest taking a closer look at the battle between WD and IBM only:

You can clearly see that every new model by Western Digital is faster than the previous one. However, IBM also makes very fast solutions :)

In this pattern we can see WD1200JB defeated by WD1000BB, a HDD with lower data density (and who told you that high data density improves the HDD performance in case of random requests?) and standard cache buffer (2MB).

DataBase pattern differs from the two previous ones by greater number of writes carried out: 33% of the entire number of operations, which lets WD1200JB get somewhat faster. WD1200JB wins all the laurels in case of lower workloads, but as soon as the queue depth increases, IBM 120GXP takes the lead.

How could this happen? We have expected that in a pattern with many write requests the fourfold advantage in the cache buffer size of WD1200JB will turn into something more tangible. Lazy Write is one of the most efficient ways to increase the HDD performance (see our article called "IBM Deskstar 60GXP: Performance Secrets") and theoretically larger cache-buffer should let the HDD postpone more writes until better times.

In order to study the behavior of the WD1200JB HDD firmware in case of many write operations, we carried out an experiment.

We created a pattern, which used absolutely random 8KB blocks of data, and the read to write ratio varied from 100 : 0 up to 0 : 100 with the step equal to 10. This way, we will be able to estimate how the HDD performance depends on this ratio. The results of this experiment turned out very interesting:

When the queue equaled 1-4 requests we could see consistent performance growth as the number of write requests increases. Note that in case of random writes the HDD runs nearly twice as fast as in case of random reads.

However, as the queue reaches 16 requests, the performance hardly gets any higher as the number of writes increases, so that the graph turns almost linear. In case of this queue depth, the HDD firmware can change the requests order and create the most optimal route for the heads, which will speed up the average requests processing. If we take a closer look at the picture, we will see that the hard drive speed drops a little bit when the number of writes reaches at least 10%. This is quite natural state of things, we should say, as any requests processing requires some time. As the number of writes grows up, the lazy write feature appears more and more profitable, so that it makes up for the losses of processor time (here we mean the DSP of the hard drive, not the CPU) required to sort all requests and to process the lists of those requests, which have been laid aside.

As the queue depth increases further on, the time required for requests processing in combo-modes (that is when there are both: reads and writes) also increases and the graph gets somewhat cup-shaped. The more mixed are the requests to be processed, the harder should the HDD work.

And now we will try to compare two hard disk drives in this pattern: WD1200JB and IBM 120GXP (we were really anxious to compare Wd1200JB with WD1200BB, but unfortunately, we failed to find the latter so far).

For a more illustrative picture we built the graphs for three workloads: 1, 16 and 256 requests.

You can clearly see that when the workload is linear, WD hard drive is just a bit slower than IBM 120GXP in those modes when the writes lie below 20%. As soon as the amount of writes exceeds 20%, WD1200JB dashes forward. The higher gets the amount of write operations, the greater turns the advantage of Wd1200JB over IBM 120GXP.

Another illustrative example is the case for 16 requests.

IBM leads while the amount of writes is relatively low, however, as the number of writes starts growing, its performance drops. It seems to be lacking the cache buffer capacity to store all "inconvenient" write requests and thus they should be processed in some inconvenient time. In fact, IBM 120GXP's firmware can assign up to 56 buffer segments to writes, but this amount seems to be too small for our case…

WD1200JB hard drive doesn't suffer any problems of the kind, its performance grows little by little as the number of writes grows.

In case of maximum workload, both HDDs have similar graphs. Moreover, IBM 120GXP doesn't suffer sharp performance drop any more, because the queue depth increased and the hard drive firmware can now better arrange the commands processing.

Temperature

Since DTemp utility reported that WD1200JB doesn't support temperature monitoring via SMART, we measured the temperature with the help of an remote infra-red thermometer. All the measurements were carried out after 2.5 hours of Intel IOMeter running. In this test the HDD works as actuator very intensively, so that it can be considered working at the top of its power.

The temperature we obtained equaled to 44oC, which is a really good result for a three-platter HDD with 7,200rpm spindle rotation speed (the room temperature was around 22oC).

Conclusion

Having given up SCSI business a couple of years ago, WD has finally took a long breath. However, the competition in the IDE market is also quite tense, and WD does its best not just to survive there, but to oust the competitors from the pedestal.

One of the attempts undertaken in this direction is the introduction of a new sub-family of IDE drives with high storage capacity and cache buffer. These solutions appear direct competitors to SCSI drives with 7,200rpm spindle rotation speed, which are actively used in High-End workstations. Moreover, these HDDs boast very attractive price-to-capacity ratio. Contemporary SCSI drives can't boast anything competitive here, even if we disregard the cost of the controller cards.

WD1200JB hard disk drive we have just reviewed proved very fast in business tests, having improved the performance top for all IDE HDDs. Moreover, WD1200JB competes on equal terms even with SCSI products rotating at 10,000-15,000rpm (see our Hard Disk Drives in 2001: Annual Overview).

In Intel IOMeter tests, to be more exact in those patterns we considered, WD1200JB didn't show any outstanding results, however, its performance in case of low workloads makes it an indisputable leader among HDDs with 40GB platters.

If you remember, in the very beginning of this review we asked ourselves a question: does it make real sense to equip an IDE hard drive with larger cache buffer, or is it just a marketing trick? So, now we have every reason to state with all certainty that large cache-buffer can push the hard disk drives with IDE interface to a new much higher performance level. Since this performance growth can be achieved at the expense of relatively low production cost increase, it will undoubtedly receive a very warm welcome. If all other hard drive manufacturers support this good start of WD's, we will have to be twice as grateful to Western Digital for this revolutionary move. First, for its great drives and second, for destroying the old stereotypes.

The interesting thing is that the price of WD1200JB is quite reasonable: around $280, which makes some $2.33 for 1GB. So, this hard disk drive represents not only a beautiful combination of great storage capacity and high speed but also costs reasonable money.

P.S.: We wonder if WD will apply the same approach (8MB cache buffer) to those HDDs, which storage capacity is lower than 100GB…?

P.P.S.: We wonder what reaction will follow from the competitors…? :)